Patterns of behaviour in nursing staff actioning the afferent limb of the rapid response system (RRS): A focused ethnography

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2020 The Authors. Journal of Advanced Nursing published by John Wiley & Sons Ltd 1School of Health Sciences, City University of London, London, UK 2University College London Hospitals NHS Foundation Trust, London, UK 3School of Nursing and Midwifery, Griffith University, Nathan, QLD, Australia


| INTRODUC TI ON
Sub-optimal care of the deteriorating ward patient was first reported in the academic literature over 20 years ago (McQuillan et al., 1998).
Sub-optimal care is a complex and multi-faceted concept that has no absolute definition. However, in the context of the deteriorating patient, key characteristics have been reported as delays in diagnosis, treatment or referral, poor assessment and/or inappropriate or inadequate treatment (Quirke, Coombs, & McEldowney, 2011).
To improve responses to deteriorating patients and mitigate the risk of sub-optimal care, acute hospitals have implemented rapid response systems (RRS) in the UK, North America, and Australasia (DeVita et al., 2006;Johnstone, Rattray, & Myers, 2007;Lyons, Edelson, & Churpek, 2018). While there is international and inter organizational variance in the operational characteristics of these systems, RRS are broadly seen to include an afferent (detection) limb and an efferent (response) limb. Behaviours of the afferent limb typically include the routine monitoring of vital signs, identification of physiological abnormality, escalation to an appropriate responder (e.g., a doctor or specialist nurse) and a subsequent increase in the frequency of monitoring (DeVita et al., 2006;Lyons et al., 2018). The afferent limb is modelled on evidence that at least 60% of patients who deteriorate in hospital have antecedent changes in vital signs preceding SAE (Andersen et al., 2016;Kause et al., 2004). Efferent limb behaviours (enacted by the responder) include further assessment, initiation of treatment or stabilizing interventions, and facilitation of patient transfer to a higher-care setting, for example, a critical care unit (Bannard-Smith et al., 2016;DeVita et al., 2006). In this work, the behaviours of interest were those of the afferent limb.
Given the relatively high frequency of premonitory signs in deteriorating patients, the use of "track and trigger" tools is recom- Healthcare Improvement (n.d.)). Broadly, "track and trigger" is a universal term describing a tool (either paper-based or electronic) on which vital signs are recorded. The tool provides a signal to clinical staff when the vital signs fall outside of acceptable parameters and then prompts staff to follow an escalation protocol (Grant, 2018).
Historically, different tools have been used creating inconsistency within and between organizations (Jansen & Cuthbertson, 2010;Shiloh, Lominadze, Gong, & Savel, 2016). To standardize UK practice, the National Early Warning Score (NEWS) was developed, published (Royal College of Physicians, 2012) and subsequently revised as NEWS2 (Royal College of Physicians, 2017). NEWS2 signals patient risk on the basis of the total score (total score range 0-20) aggregated from individual scores assigned to six routinely recorded vital signs (Table 1). The patient's risk level is then stratified according to the aggregate score (File S1) which aligns to an associated escalation algorithm (Royal College of Physicians, 2017).
Despite international implementation of RRS and the availability of NEWS2 in the UK, there is evidence that staff do not change their behaviour to increase the frequency of vital signs monitoring or escalate care when criteria are met (Credland, Dyson, & Johnson, 2018).
This lack of compliance has been termed "afferent limb failure" (ALF) (Johnston, Arora, King, Stroman, & Darzi, 2014;Trinkle & Flabouris, 2011). There is an abundance of literature describing the potential causes of ALF (Olsen, Søreide, Hillman, & Hansen, 2019;Treacy & Stayt, 2019;Wood, Chaboyer, & Carr, 2019) but paucity of work reporting interventions to target it (Bucknall et al., 2017;Connell et al., 2016;Duff, Massey, Gooch, & Wallis, 2018). Further, most of the interventions described are educational with methodological limitations including risks of bias and/or consistently poor detailing of the development process, suggesting that these interventions may have been developed pragmatically (i.e., based on clinician or researcher intuition) rather than using a replicable method.
Given its pervasive nature, there is an argument for using more systematic behavioural approaches to investigate and address ALF.

| BACKG ROU N D
While several different approaches for developing interventions are reported, we used a theory-based approach for intervention  (Medical Research Council, 2006). The application of theory enables determinants (i.e., barriers and enablers) of behaviour to be identified and for intervention components that specifically target these determinants to be selected and tailored to context (Cadogan et al., 2015;Patton et al., 2018). To develop a theory-based intervention, clear specification of target behaviours is required to enable measurement of behaviour change in subsequent intervention testing (Atkins et al., 2017;Presseau et al., 2019). Before reporting undesirable or deviant behaviours (i.e., those that will be targeted by the intervention), expected behaviour must first be specified.

| Aim and objectives
The aim of this project was to improve understanding of afferent limb behaviour in an acute hospital ward setting. Specific objectives were: • To compare expected (i.e., policy specified) behaviours of nursing staff with those observed on hospital wards • To report where afferent limb failure was occurring in the sequence of observed behaviours • To define and specify the behaviours that could be targeted by a theory-based intervention, using the five criteria of a published behaviour specification framework -action, actor, context, target, and timing .

| Design
This project is one component of a multi-phase intervention development process, for which a protocol has been published . Focused ethnography was conducted to explore the behaviour of nursing staff, working in acute hospital wards, when they were actioning behaviours of the afferent limb of the RRS. Focused ethnography is an applied qualitative methodology that is well suited to research where participants reflect a small sub-group of society (e.g., a particular professional group), where the objective is to elucidate a reported problem in a particular context, and where the researcher's access to participants is limited to brief, episodic contact (Cruz & Higginbottom, 2013;Knoblauch, 2005).

| Sample
This research was conducted in an acute metropolitan hospital in England that provides care for the local population as well as specialist services. The organization comprises seven geographically separate hospitals with a total bed-base of 1,161. This study was conducted in the largest in-patient site in the organization. In 2018, it was confirmed that the hospital would be switching from paperbased patient records to an Electronic Health Record System (EHRS).
Part of this process was migration from a paper-based NEWS chart to an electronic version of NEWS2. These plans were announced after this study had been designed and the original protocol written.
It was identified that this period of transition would provide a unique opportunity for data collection before and after the implementation The other area (floor A) had no such investigations in progress.
Further characteristics of the clinical floors are described in Table 2.
Based on methodological precedent (Mackintosh, Humphrey, & Sandall, 2014), we proposed to observe for 180 hr on different days of the week and at different times of day and night, or until data saturation was achieved (i.e., no new behaviours were seen). A purposive sample (balance of clinical banding) of nurses enacting behaviours of the afferent limb were observed.

| Data collection
Data collection activities were conducted in two phases. Between 7 January 2019 -27 March 2019, data were collected in the paperbased context. Between 1 January 2019 -18 December 2019 data were collected in the electronic context. We acknowledged that staff behaviour immediately after EHRS implementation was unlikely to reflect usual practice. As such, an acclimation period of 3 months (Bedoya et al., 2019) was allowed when no data were collected.
In keeping with the concept of focused ethnography, observa- reported methods (Smith, Sekhon, et al., 2019) and guided by a behaviour specification framework . The guide focused observation on five key moments of the afferent limb (Table 3).
In conjunction with the observation guide, a document for recording field notes was developed (File S3). This document was structured to enable descriptions of staff behaviour to be recorded in addition to data from vital signs charts (paper and electronic). The document also provided a space for "reflexive notes," that is, a space for the researcher to record thoughts, feelings and interpretations of events.
The researcher conducted NEWS chart reviews throughout the data collection period. Individual vital signs and aggregate scores from the NEWS chart were extracted and recorded in the field notes. Chart review was frequently performed alongside, or in response to, direct observation. A chart review was also performed if the researcher overheard discussions about an unwell patient at nursing staff handover or a "huddle," or if the researcher observed "heightened activity" around a particular patient (e.g., staff bringing emergency equipment to the bedside).
In adult patients hospitalized for a range of clinical diagnoses respiratory rate was found to be an independent predictor of adverse events (Escobar et al., 2012;Fieselmann, Hendryx, Helms, & Wakefield, 1993;Fine et al., 1997). Unlike the other vital signs entered into NEWS, the respiratory rate is typically not measured using electronic equipment and must be measured visually by a healthcare provider (Badawy, Nguyen, Clark, Halm, & Makam, 2017 intrusive to the patient. These measurements were taken on an ad hoc basis, typically alongside direct observation and chart review. Where the researcher respiratory rate was considerably different to the recorded respiratory rate (i.e., different enough to change the NEWS risk level), an agreed safety algorithm was followed to safeguard the patient. This algorithm prompted the researcher (DS) to take a stepwise series of actions beginning with notification of the responsible RN, followed by escalation to the nurse in charge of the ward, followed by, if necessary, a call to the medical team or CCOT. The response was proportionate to the degree of physiological abnormality (i.e., how high the NEWS or how deranged the vital signs) and also the appropriateness of the observed response from the ward-based nursing staff (e.g., if the researcher prompted the RN to take action and they appeared to enact the policy specified behaviour, then no further escalation was taken by DS). Further detail of this escalation algorithm can be found in the study protocol . We received ethical approval to use an "opt-out" consent approach for this research, meaning that nursing staff were provided with multiple opportunities to opt out of participating in the study. At the beginning of a shift where DS was present, staff were reminded that they should declare (verbally or in written form) if they did not wish to be observed or approached during the period of observation. These staff were asked to prospectively sign an opt-out form.

| Ethical considerations
Copies of the opt-out form were also left in the staff room along with a sealed box so that staff could privately complete and return an opt-out form, if they did not wish to approach DS in person.
Staff who opted out were not required to specify their reasons for doing so. The completed opt-out form allowed DS to identify staff on duty who did not wish to participate (by cross-checking with the roster and staff allocation board) so that no further information was collected from these individuals. Further details of the consent procedures can be found in the study protocol .

| Data analysis
One member of the research team (DS) used structured content analysis (Hsieh & Shannon, 2005) to analyse field notes as follows: • Handwritten descriptions of direct observations and chart review data were read superficially and then more thoroughly to ensure familiarization with the subject matter.
• Data were initially labelled and categorized by the five key moments of the afferent limb.
• Within each of the five categories, data were examined further and compared directly to policy-specified behaviour (obtained from documentary analysis). If the observational data, or information extracted from chart review, aligned to policy-specified behaviour, this was categorized as "expected behaviour." Where the recorded data did not align to the policy-specified behaviour, it was categorized as "unexpected behaviour." A lack of action was also categorized as "unexpected behaviour." • Where the extracted data included a researcher respiratory rate measurement alongside a recorded respiratory rate, a sub-analysis was performed by comparing the two respiratory rate measurements. If the difference between the two measurements was greater than 5, or the difference was sufficient to change the aggregate NEWS, the episode was categorized as "unexpected behaviour." If these criteria were not met, the episode was categorized as "expected behaviour." The difference between the researcher respiratory rate and the recorded respiratory rate was summarized descriptively.
• Frequencies and proportions of expected and unexpected behaviours were counted across the corpus of data and for each of the key moments of the afferent limb.
• Unexpected behaviours were scrutinized and statements describing "who needs to do what differently" were synthesized and structured using the AACTT framework  to report target behaviours for a behaviour change intervention (to be reported in a subsequent paper).

| Rigour
In qualitative research, multiple data collection strategies may be reported methods (Cruz & Higginbottom, 2013). Both these approaches were incorporated into this design.
Comprehensive field and reflexive notes were taken throughout the period of data collection to ensure dependability of the

| FINDING S
Across the two clinical floors, a total of 300 hr of observation was carried out; 150 hr when a paper-based NEWS chart was in use (i.e., pre EHRS implementation) and 150 hr when an electronic NEWS2 chart was in use (i.e., post EHRS implementation) (Figure 1 shows a detailed breakdown of these hours by floor). Four members of staff (all HCAs) prospectively opted-out of being observed (staff were not required to declare why they chose to opt out).
Four hundred and ninety-nine discrete items of data (e.g., a single episode of observational data, or a single set of vital signs from one occurrence of patient monitoring) were extracted from field notes and analysed; 253 items of data were collected pre EHRS; 246 items of data were collected post EHRS. Two hundred and eighty-nine (58%) items of data were associated with expected (e.g., policy-specified) afferent limb behaviour; 210 (42%) items of data were associated with unexpected afferent limb behaviour (e.g., alternative behaviour or no behaviour) ( Table 4 displays the frequency of expected and unexpected behaviour for each of the five key moments of the afferent limb). Ten specific behaviours were identified where the behaviour observed deviated (negatively) from policy or where no action was taken when it should have been (these potential targets for behaviour change are described in Table 5). One further behaviour was seen to expedite the assessment of a deteriorating patient by an appropriate responder and was therefore considered a positive deviant behaviour. Descriptive accounts of field data are reported below in relation to each key of the key moments of the afferent limb. File S4 contains exerts extracted directly from field notes in support of each of these accounts. In some cases, it was very clear that the HCA or RN being observed were enacting expected behaviour in counting the patient's respiratory rate as part of routine monitoring. In these instances, staff were seen looking at a fob watch on their uniform, at a wallmounted clock or, more frequently, at a timer on an electronic ther- stating to a colleague that they did not have sight of a clock. Despite this, they proceeded to record a respiratory rate on the NEWS chart.

| Routine monitoring of vital signs
Some staff were also seen enacting unexpected behaviour in relation to the use of electronic monitoring equipment. On several occasions, HCAs were observed applying finger probes for measuring Sp0 2 to a patient's ear. This was often seen in response to the monitoring equipment alarming when first applied to a digit.

| Responsive monitoring of vital signs
The expected behaviour of responsive monitoring typically involved the monitoring of vital signs in a single patient more frequently than other patients in their bay. Both RNs and HCAs were seen enacting these behaviours in the pre and post EHRS context. RNs were more frequently observed enacting responsive monitoring compared with routine monitoring. On some occasions, electronic monitoring devices were left connected to the patient and stationed in the patient's bed space to permit more frequent measurement of vital signs. This was recorded as expected behaviour in the context of a deteriorating patient.
When approached by a HCA about a patient with an elevated NEWS or abnormal vital signs, RNs were seen to delegate further monitoring back to a HCA or student nurse, rather than assessing the patient further themselves (the expected behaviour). This was observed on multiple occasions involving different patients including a patient with an un-recordable blood pressure, a patient who had already been reviewed by critical care and a patient with a high NEWS.
Chart reviews were frequently conducted to assess the timeliness of repeat monitoring after a NEWS trigger. Examples of expected behaviour were found illustrating monitoring frequency being increased, according to policy, for medium and high-risk NEWS. There was also evidence of unexpected behaviour in view of delayed monitoring (i.e.,> 1 hr between episodes) for patients with both medium and high-risk scores.

| Recording vital signs and/or calculating the NEWS
The behaviours related to the recording of vital signs and the generation of an aggregate NEWS, were the most variable between the pre and post EHRS periods. In the pre EHRS context, review of paper NEWS charts highlighted inconsistency in the accuracy of recorded information. On some occasions, evidence of expected behaviour was found whereby all vital signs were recorded legibly and an accurate NEWS was calculated. On other occasions, specific vital signs were missing, or an aggregate NEWS was not recorded, or the aggregate NEWS was recorded but was not calculated correctly. Infrequently, the time recorded on the NEWS tool (paper and electronic) appeared to reflect the time that the vital signs were due rather than the time that they were seen to be measured. This was considered unexpected behaviour.
The EHRS appeared to remedy errors in the calculation of NEWS, however, there were still occasions where incomplete recording of vital signs by staff (unexpected behaviour) prevented the TA B L E 5 Description of policy-practice gaps and specification of afferent limb behaviours that could be targeted by a theory-based behaviour change intervention

Who needs to do what differently (potential target for the behaviour change intervention)
Every time an HCA/RN measures vital signs, all 6 parameters should be recorded, and an accurate NEWS calculated (this is automated on the EHRS) HCAs were observed writing vital signs on a piece of paper or handover sheet or paper towel, and were later were seen entering a whole bay/group of patients' vital signs into NEWS EHRS generating an aggregate score. Also, where patients were visibly confused/delirious, this was not always recorded and scored as expected on the NEWS chart.
The practices of staff when recording the vital signs was highly variable. In the post EHRS context, some HCAs and RNs were seen to enter vital signs directly into either a desktop computer or a workstation on wheels. Some HCAs used hand-held devices to enter the vital signs immediately after they had measured them. All these behaviours facilitated contemporaneous recording and were therefore considered expected. Other HCAs were observed jotting several patients' vital signs down on a piece of paper (typically a paper towel or clinical handover sheet) before then entering them into the EHRS later using a desktop computer. These behaviours created a delay in recording and were therefore considered unexpected.

| Differences between recorded respiratory rate and researcher respiratory rate
On 37 occasions (across the pre and post EHRS data collection periods), a researcher respiratory rate was counted and compared with values recorded by HCAs and RNs. The median difference between the recorded respiratory rate and researcher respiratory rate was 5 (IQR 1-10). In 28 (76%) cases, the researcher respiratory rate was higher than the recorded respiratory rate. In 24 (65%) cases, the researcher's calculated NEWS was higher than the recorded NEWS; in 17 (46%) cases, the researcher NEWS resulted in an upgrade of the NEWS risk level and therefore a different recommended course of action. In 10 (27%) cases, the level of risk would have been upgraded to either medium (19%) or high (8%) risk, from a lower-risk category.

| Escalation within the ward-based nursing team
In both the pre and post EHRS context, escalation behaviours were less frequently observed than monitoring, recording, and scoring behaviours. HCAs were observed escalating, as expected, to RNs in the pre and post EHRS contexts and were typically overheard reporting concerns with specific vital signs. Less frequently, HCAs were overheard raising concerns about an elevated NEWS. However, on both floors, there were situations where patients with abnormal vital signs and elevated NEWS had not been escalated, as expected, by the HCA who undertook the measurements to the responsible RN.

| Escalation outside of the ward-based nursing team
On both floors, RNs were observed escalating, as expected, to external personnel including medical staff and CCOT. These behaviours were enacted in both the pre and post EHRS contexts. In most cases, the escalation occurred via the hospital pager system, which

Who needs to do what differently (potential target for the behaviour change intervention)
After recording a NEWS ≥5, the RN should escalate to the parent medical team +/-CCOT +/-night nurse practitioners RNs do not consistently escalate patients with elevated NEWS. This includes patients under CCOT and/or those flagged as 'at risk' (at safety huddles etc.) Paper Escalation (action) to the parent medical team and/or CCOT and/or night nurse practitioners (primary targets) should be carried out by RNs (actor) when NEWS is ≥5 (context), in any ward patient (secondary target), after they have re-measured vital signs and/or completed an ABCDE assessment (timing) unless a reasonable variance has been agreed and documented.
If the first responder to whom the RN escalates does not respond as expected, then the RN contacts other personnel (e.g., a different doctor or CCOT nurse) to ensure that the patient is assessed and/or a clear plan is made EHRS Further escalation (action) to second responder (e.g., a different doctor or CCOT nurse) (primary target) should be carried out by a RN (actor), if the first practitioner they approached cannot attend or does not respond as policy states, during any episode of escalation to any responder (context) at any time of day or night (timing).
After recording a NEWS ≥5, the frequency of vital signs monitoring should be increased to a minimum of 1 hourly measurements The primary target(s) of the specified behaviour are the individual(s)/group(s) who must decide whether subsequent behaviours are required, while the secondary target(s) are the individual(s)/group(s) who benefit from the specified behaviour but are not required to enact anything themselves.

TA B L E 5 (Continued)
involved staff dialling a pager number into the telephone, entering their contact extension for the responder and then waiting by the telephone for the responder to return their call. On floor A, there were several occasions where escalation to medical staff occurred in person rather than over the telephone. Typically, this involved an RN approaching a doctor from the office on the ward and bringing them to the bedside of a patient.
There were instances where patients met the criteria for escalation but had not been escalated by RNs to the CCOT. One example of this unexpected behaviour involved a patient who had already been identified as potentially needing a step-up of care to ICU, who was not escalated in response to an elevated NEWS.

| D ISCUSS I ON
During the period of observation, expected and unexpected behaviours were observed in four of the five key moments of the afferent limb in both the paper and EHRS contexts. For the key moment of "escalation outside of the ward-based nursing team," only expected (policy-specified) behaviour was observed in the EHRS context.
More than 90% of the data collected related to monitoring, recording and, when required, scoring behaviours. Less than 10% of the data collected reflected behaviours of escalation. There were clear areas of "role overlap" where the expected behaviour was enacted by both RNs and HCAs, particularly in responsive monitoring of vital signs. Other behaviours were more delineated by role. In particular, routine monitoring of vital signs was nearly always enacted by HCAs.
Conversely, higher level escalations (i.e., outside of the ward-based nursing team) were exclusively actioned by RNs. Some unexpected behaviours involved actions that deviated from policy or practice guidelines. In these cases, the behaviour was broadly enacted but not to the standard of best practice, for example, monitoring vital signs but misusing equipment (e.g., applying a pulse oximetry probe designed to be applied to a patient's finger, to the ear). More commonly, unexpected behaviour involved no action, for example, an RN not escalating an elevated NEWS to CCOT.
Most routine monitoring of vital signs involved the use of electronic monitoring equipment and was typically performed by HCAs. These findings are consistent with other literature (Ede, Jeffs, Vollam, & Watkinson, 2019;Mackintosh et al., 2014;Smith & Aitken, 2016) implying this may be common practice. There were exceptions where RNs were seen undertaking routine monitoring, this typically occurred in the context of short staffing or when a HCA was re-deployed to a "heavier" part of the ward. The assumption that HCAs will undertake what Ede et al. (2019) describe as "bulk monitoring" (p4) presents several potential challenges. First, it establishes a disconnect within the afferent limb between the actor responsible for collecting the clinical data, that is, measuring the vital signs and the actor expected to evaluate the information and act (Mackintosh et al., 2014). Arguably, it also denies the RN a further opportunity to interact with the patient and capture additional clinical information (Cardona-Morrell et al., 2015). In the context of patient deterioration, there is evidence that "nurse worry" is important in predicting adverse patient outcomes (Douw, Huisman-de Waal, van Zanten, van der Hoeven, & Schoonhoven, 2016;Romero-Brufau et al., 2019). While "nurse worry" has been linked to tacit knowledge, it may also arise from a more comprehensive assessment and the collection of additional clinical cues (e.g., patient appearing agitated or skin clammy to touch) (Douw et al., 2015). Through undertaking routine monitoring, RNs would be well positioned to identify these additional cues alongside the vital signs. At present, there is no information in the published literature about HCA worry, including whether or not HCAs are sensitive to the same cues of deterioration as RNs, or if their sense of worry has predictive validity.
In view of this, deteriorating patient policies typically stipulate that HCAs, carrying out routine monitoring, should have a low threshold to escalate if the NEWS is elevated or vital signs abnormal (Smith, Sekhon, et al., 2019). Despite this, situations were observed where HCAs had measured and recorded an elevated NEWS but not notified the RN. The lack of expected behaviour from the HCA created a "hard stop" in the sequence (i.e., no further action taken), as the RN behaviours were contingent on activation from the HCA. Some authors have argued that increasing reliance on un-registered staff to undertake safety-critical aspects of nursing, reflects a wider challenge facing the workforce where RN expertise is increasingly devalued and diluted (Leary, 2019). This is particularly concerning, given the evidence that adverse outcomes are reduced when patients are cared for in organizations with higher numbers of well-educated registrants (Aiken et al., 2011).
Using focused ethnography, we identified unexpected behaviour in the monitoring and recording of patients' respiratory rate by HCAs and RNs. In three quarters of cases, the observed respiratory rate by the researcher was higher than the respiratory rate recorded on the chart and, in almost half of the cases, the NEWS would have been higher if the recorded respiratory rate was replaced with the researcher respiratory rate. Our finding that respiratory rate is often under reported, leading to a potential underestimation of patient acuity, is consistent with other research including a study that compared respiratory rate measured by an electronic wearable device to respiratory rate measured by nurses (Weenk et al., 2019). Cited explanations for this unexpected behaviour include a lack of skill in obtaining the measurement and a lack of knowledge of its importance (Treacy & Stayt, 2019). Use of wearable continuous respiratory rate monitoring devices offer one solution to this pervasive problem (Weenk et al., 2019). However, a targeted intervention to ensure more consistent staff behaviour in this area could be a feasible alternative.
As the use of technology in healthcare becomes increasingly pervasive, interest has grown on the impact of technology on patient safety, more specifically on its impact on the RRS (Wilson & Khansa, 2018). In the paper context, errors have been reported in the recording and calculation of aggregate early warning scores (EWS) often leading to an under estimation of patient risk and sub-optimal responses (Kolic, Crane, McCartney, Perkins, & Taylor, 2015;Odell, 2015). Our findings broadly corroborate these reports, as more than 40% of observed recording and scoring behaviours were categorized as unexpected in the paper NEWS context.
Comparatively, there is evidence that scoring automation within an EHRS-embedded EWS completely eliminates human error in score calculation (Credland et al., 2018;Jones et al., 2011). However, in some EHRSs, the healthcare provider is still required to manually key the data into the system. We observed cases where the NEWS was not calculated by the EHRS due to missing or inaccurately entered data. Our findings align to other published literature, also reporting the problem of incomplete vital signs in an EHRS context (Stevenson, Israelsson, Nilsson, Petersson, & Bath, 2016). This unexpected behaviour could be the result of a lack of knowledge among nursing staff about the importance of an aggregate NEWS in determining risk, or lack of awareness of the potential consequences of not completing a thorough and timely patient assessment (Treacy & Stayt, 2019;Wood et al., 2019).
In view of unexpected behaviour in the recording and scoring key moment, we also observed staff (predominantly HCAs) writing a series of vital signs on paper before, then entering them all into the EHRS. This appeared to delay the availability of the data to other members of the healthcare team (including the RN), delayed the generation of a NEWS and led to transcription error. While this unexpected behaviour was seen in both the paper and EHRS contexts, the frequency of this specific behaviour increased after implementation of the EHRS. These behaviours, described in the literature as use of paper "workarounds" (Stevenson, Israelsson, Petersson, & Bath, 2018), have been attributed to dissatisfaction of staff with the layout and presentation of vital signs on the EHRS and a lack of equipment to enter the data, leading them to enact alternative behaviours (Stevenson et al., 2016(Stevenson et al., , 2018. What is clear, is that the implementation of the EHRS is not a panacea for ALF. While some negative deviant behaviours are reconciled, others may increase suggesting these systems may have the potential to improve patient safety (Jones et al., 2011), however, careful consideration of the environmental and behavioural context is required.
We elected to collect data before and after the implementation of an EHRS to maximize researcher exposure to different behaviours of the afferent limb. While our study was not designed to signal cause and effect of EHRS implementation, it is noteworthy that 6 of 10 negative deviant behaviours were observed in both the pre and post EHRS contexts, suggesting these behaviours may be deeply entrenched. Further, in light of evidence that habit plays a significant role in health professional behaviour (Potthoff et al., 2019), it is plausible that some of these behaviours are enacted automatically, rather than based on careful and deliberative reasoning (Presseau et al., 2014). If this is the case, carefully selected and tailored intervention components will be required to change staff behaviour.

| Strengths and Limitations
In the context of the deteriorating hospital patient, we believe that this is the first paper to report, comprehensively, the use of focused ethnography to describe and specify behaviours that could be targeted by a theory-based implementation intervention. In the wider behaviour change literature, researchers have used local audit to identify who needs to change their behaviour (Taylor et al., 2016).
While an acceptable approach, there is arguably a risk that some of the more nuanced and context-specific aspects of behaviour may not be captured. By comparison, focused ethnography has the potential to provide deeper insight into behaviour as it occurs within the "natural setting" (Leslie,  Our procedure for the counting and comparing of respiratory rates had inherent limitations. First, it is plausible that the respiratory rate may have changed in the period (maximum 15 min) between it being recorded by the nurse/HCA and the researcher. In these circumstances, the behaviour may have been reported as unexpected when, in fact, the change was physiological rather than "user error," that is, a miscount by the RN or HCA. Further, it is possible that the "user error" belonged to the researcher rather than the ward staff. However, the researcher is an experienced RN with expertise in clinical assessment, specifically the assessment of deteriorating patients. In addition, the researcher was arguably less likely to be distracted by other activity on the ward and was able to repeatedly measure the respiratory rate, over a full minute, until he felt confident in the measurement.

| CON CLUS ION
Using focused ethnography, we identified and specified 10 deviant afferent limb behaviours that could be targeted for change and a further behaviour that could be enabled, by a theory-based implementation intervention. Five of these behaviours were only observed in the pre or post EHRS context. However, it is possible that these behaviours were enacted in both settings but not detected by the observer. As such, all 11 specified behaviours could be considered as potential intervention targets. Further theory-based inquiry is required to elucidate the determinants of these behaviours, to map these determinants to intervention components and tailor the delivery to context.

ACK N OWLED G EM ENTS
The authors would like to acknowledge Professor Jill J Francis, who played an integral role in designing the study within which this research is situated.

CO N FLI C T O F I NTE R E S T
No conflict of interest has been declared by the author(s).

AUTH O R CO NTR I B UTI O N S
All authors have agreed on the final version and meet at least one of the following criteria (recommended by the ICMJE (http://www. icmje.org/recom menda tions/): (1) substantial contributions to conception and design, acquisition of data, or analysis and interpretation of data; (2) drafting the article or revising it critically for important intellectual content.

PE E R R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/jan.14551.